DOCSLIB.ORG

An Overview of the Crúbadán Project

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
An Overview of the Crúbadán Project An Overview of the Crúbadán Project Kevin Scannell Saint Louis University 11 August 2014 Background ● I was born in the US, but I speak Irish ● Trained as a mathematician ● Began developing tools for Irish in the 90s ● No corpora for doing statistical NLP ● So I built my own with a crawler ● Crúbadán = crawling thing, from crúb=paw An Crúbadán: History ● First attempt at crawling Irish web, Jan 1999 ● 50M words of Welsh for historical dict., 2004 ● ~150 minority languages, 2004-2007 ● ~450 languages for WAC3, 2007 ● Unfunded through 2011 ● Search for “all” languages, started c. 2011 How many languages on the web? ● Two year project ends in December ● Phase one: aggressively seek out new langs ● Phase two: produce free+usable resources ● Current total: 1950 ● At least 150 more queued for training ● 2500? 3000? Languages vs. time Goals ● Interested in revitalization, first and foremost ● Building useful software for communities ● Blark: word lists, morph. analyzers ● Open data for under-resourced languages ● Linguistic typology ● Linguistic diversity of the web Spelling and grammar checkers ● Corpus-based Irish spellchecker, 2000 ● Grammar checker, 2003 ● 28 new spellcheckers since 2004 ● Collaborations with native speakers ● All under open source licenses hunspell ● Standard open source spellchecking engine ● The default in Firefox and LibreOffice ● Fast and powerful ● Good language support: ~150 languages ● Can be as simple as a word list ● But also supports complex morphology Morphology in hunspell ● Finite-state transducers (Xerox, HFST, …) ● Very fast + bidirectional ● Cover most morphology of human langs ● Hunspell uses “two-fold affix stripping” ● Morphological analysis only ● Not as powerful, theoretically ● BUT: simple formalism, user support FTW! Powerful enough ● Hungarian ● Northern Sámi ● Basque ● Lingala, Kinyarwanda, Swahili, Chichewa ● …? Language ID ● Component and an application of Crúbadán ● Character n-grams + word models ● NLTK 3-gram data set ● Indigenous Tweets and Blogs Predictive text ● T9 input ● Adaptxt ● Firefox OS accentuate.us ● Diacritic restoration service ● Eni kookan lo ni eto si omi nira lati ni imoran ti o wu u ● Ẹnì kọ̀ọ̀kan ló ní ẹ̀tọ́ sí òmì nira láti ní ìmọ̀ràn tí ó wù ú ● End-user clients for Firefox, LibreOffice ● Perl, Python, Haskell libraries ● Joint work with Michael Schade Lexicography ● Geiriadur Prifysgol Cymru ● Foclóir Nua Béarla-Gaeilge ● Foclóir na Nua-Ghaeilge ● SketchEngine NLP Research ● N-gram language models for MT ● Computational morphology ● Parsing ● OCR (e.g. Irish seanchló) ● Speech recognition/synthesis N-Gram Models ● Generative language model ● Prob of word conditioned on previous n-1 ● Used in “noisy channel” problems ● Machine Translation ● Speech recognition ● Spelling and grammar correction Irish standardization ● Irish underwent a spelling reform in the 40's ● Hard to use pre-standard texts for NLP ● Treated this as a statistical MT problem ● Just need an n-gram model for target ● Translation handled via spelling rules ● Plus hand-curated pre-post mappings Example ● Pre: Ní rabh 'sa dearbhughadh sin acht a chuid uchtaighe, eisean, a h-Aodh féin ag teacht na h-arraicis. ● Post: Ní raibh sa dearbhú sin ach a chuid uchtaí, eisean, a hAodh féin ag teacht ina haraicis. Scottish Gaelic to Irish MT ● Very similar setup ● SG spelling resembles pre-standard Irish ● Many of the same spelling rules ● Exactly the same n-gram model for target Example ● Sc: Bha tàladh air choireigin na nàdur a bha a' tarraing a h-uile duine thuice. ● Ir: Bhí mealladh éigin ina nádúr a tharraing gach uile dhuine chuici. Linguistic research ● Comparative phonology ● Syntax ● Psycholinguistics ● Selectional preferences ● … Orthotree ● http://indigenoustweets.blogspot.com/2011/12/ ● https://github.com/kscanne/orthotree An Crúbadán: Design principles ● Orthographies, not languages ● Labelled by BCP-47 codes ● en, chr, sr-Latn, de-AT, fr-x-nor, el-Latn-x-chat ● Real, running texts (vs. word lists, GILT) ● Get “everything” for small languages ● Large samples for English, French, etc. Three modules ● Traditional web crawler ● Twitter crawler ● Blog tracker Phase 1: Finding new languages ● Lots of web searching! ● Special code monitors WP, JW, UN, bible.is ● Typing/OCR of scanned or offline texts ● Special thanks to Ed Jahn of George Mason ● D. Joosten, J. Berlage, N. Lewchenko ● NSF grant 1159174 Phase 2: Building useful resources ● Separating orthographies/dialects ● Clean boilerplate ● Convert to UTF-8 text + normalize ● Sentence segment and tokenize ● Avoid copyright issues ● Discoverability (OLAC) UTF-8 Normalization ● Fonts (Sámi, Mongolian, dozens of others) ● Lookalikes (az: ә/ә, bua: γ/ү, ro: ş/ș) ● Shortcuts (haw, mi, etc. äëïöü for āēīōū) ● Encoding issues (tn, nso: ß/š from Latin-2) ● Apostrophe hell: 'ˈ´ʹʻʼʽʾˊˋ˴ʹ΄՚՝᾽᾿´῾‘’‛′‵ꞌ ＇｀ Tokenization ● Default tokenizer (letters in default script) ● Many exceptions: Greek in coo/hur/kab, etc. ● Word internal punctuation (ca: l•l, l·l) ● Initial/final apostrophes or lookalikes Twitter crawler ● Twitter's REST API ● Search “unique” words ● Crawl follow graph ● Language ID particularly challenging ● Uses word and character models Blog tracker ● Blogger platform only (for now) ● Works hand-in-hand with traditional crawler ● Registers all blogs with an in-language post ● Tracks all past and future posts ● http://indigenousblogs.com/ Call to action ● > 100 collaborators: speakers, linguists ● Help sort dialects, orthographies ● Tokenization and normalization ● Finding new material for training ● Help create new online material .
Load more

Recommended publications
  • Powering Inclusion: Artificial Intelligence and Assistive Technology
    POLICY BRIEF MARCH 2021 Powering Inclusion: Artificial Intelligence and Assistive Technology Assistive technology (AT) is broadly defined as any equipment, product or service that can make society more inclusive. Eye- glasses, hearing aids, wheelchairs or even some mobile applica- tions are all examples of AT. This briefing explores the power of Key facts using Artificial Intelligence (AI) to enhance the design of current and future AT, as well as to increase its reach. The UN Convention on the Context Rights of Persons with Dis- Globally, over 1 billion people are currently in need of AT. Lack of access to basic AT excludes individuals, reduces their ability to live independent abilities (UNCRPD) estab- lives [3], and is a barrier to the realisation of the Sustainable Development lished AT provision as a Goals (SDGs) [2]. human right [1]. Advances in AI offer the potential to develop and enhance AT and gain new insights into the scale and nature of AT needs (although the risks of potential bias and discrimination in certain AI-based tools must also be Over 1 billion people are acknowledged). In June 2019, the Conference of State Parties to the UN currently in need of AT. Convention on the Rights of Persons with Disabilities (CRPD) recognized that AI has the potential to enhance inclu- By 2050, this number is sion, participation, and independence for people with disabilities [5]. AI- predicted to double [2]. enabled tools – such as predictive text, visual recognition, automatic speech- to-text transcription, speech recognition, and virtual assistants - have experi- enced great advances in the last few years and are already being Only 10% of those who used by people with vision, hearing, mobility and learning impairments.
    [Show full text]
  • An Improved Text Sentiment Classification Model Using TF-IDF and Next Word Negation
    An Improved Text Sentiment Classification Model Using TF-IDF and Next Word Negation Bijoyan Das Sarit Chakraborty Student Member, IEEE Member, IEEE, Kolkata, India Abstract – With the rapid growth of Text sentiment tickets. Although this is a very trivial problem, text analysis, the demand for automatic classification of classification can be used in many different areas as electronic documents has increased by leaps and bound. follows: The paradigm of text classification or text mining has been the subject of many research works in recent time. Most of the consumer based companies use In this paper we propose a technique for text sentiment sentiment classification to automatically classification using term frequency- inverse document generate reports on customer feedback. It is an frequency (TF-IDF) along with Next Word Negation integral part of CRM. [2] (NWN). We have also compared the performances of In medical science, text classification is used to binary bag of words model, TF-IDF model and TF-IDF analyze and categorize reports, clinical trials, with ‘next word negation’ (TF-IDF-NWN) model for hospital records etc. [2] text classification. Our proposed model is then applied Text classification models are also used in law on three different text mining algorithms and we found on the various trial data, verdicts, court the Linear Support vector machine (LSVM) is the most transcripts etc. [2] appropriate to work with our proposed model. The Text classification models can also be used for achieved results show significant increase in accuracy Spam email classification. [7] compared to earlier methods. In this paper we have demonstrated a study on the three different techniques to build models for text I.
    [Show full text]
  • Extraction of Predictive Document Spans with Neural Attention
    SpanPredict: Extraction of Predictive Document Spans with Neural Attention Vivek Subramanian, Matthew Engelhard, Samuel Berchuck, Liqun Chen, Ricardo Henao, and Lawrence Carin Duke University {vivek.subramanian, matthew.engelhard, samuel.berchuck, liqun.chen, ricardo.henao, lcarin}@duke.edu Abstract clinical notes, for example, attributing predictions to specific note content assures clinicians that the In many natural language processing applica- tions, identifying predictive text can be as im- model is not relying on data artifacts that are not portant as the predictions themselves. When clinically meaningful or generalizable. Moreover, predicting medical diagnoses, for example, this process may illuminate previously unknown identifying predictive content in clinical notes risk factors that are described in clinical notes but not only enhances interpretability, but also al- not captured in a structured manner. Our work is lows unknown, descriptive (i.e., text-based) motivated by the problem of autism spectrum dis- risk factors to be identified. We here formal- order (ASD) diagnosis, in which many early symp- ize this problem as predictive extraction and toms are behavioral rather than physiologic, and address it using a simple mechanism based on linear attention. Our method preserves dif- are documented in clinical notes using multiple- ferentiability, allowing scalable inference via word descriptions, not individual terms. Morever, stochastic gradient descent. Further, the model extended and nuanced descriptions are important decomposes predictions into a sum of contri- in many common document classification tasks, for butions of distinct text spans. Importantly, we instance, the scoring of movie or food reviews. require only document labels, not ground-truth Identifying important spans of text is a recurring spans.
    [Show full text]
  • Alternate Keypad Designs and Novel Predictive Disambiguation Methods
    Improved Text Entry for Mobile Devices: Alternate Keypad Designs and Novel Predictive Disambiguation Methods A DISSERTATION SUBMITTED TO THE COLLEGE OF COMPUTER SCIENCE OF NORTHEASTERN UNIVERSITY IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY By Jun Gong October 2007 ©Jun Gong, 2007 ALL RIGHTS RESERVED Acknowledgements I would like to take this chance to truly thank all the people who have supported me throughout my years as a graduate student. Without their help, I do not think I could have achieved one of my life-long goals and dreams of earning a doctorate, and would not be where I am today. First of all, I have to express my sincerest appreciation to my advisors Peter Tarasewich and Harriet Fell. I shall never forget these four years when I worked with such distinguished researchers and mentors. Thank you for always putting your trust and belief in me, and for your guidance and insights into my research. I also have to thank my other thesis committee members: Professors Scott MacKenzie, Carole Hafner, and Javed Aslam. I would not be where I am now without the time and effort you spent helping me complete my dissertation work. I am most indebted to my beloved parents, Pengchao Gong and Baolan Xia. I have been away from you for such a long time, and could not always be there when I should have. But in return, you have still given me the warmest care and greatest encouragement. I am forever grateful that you have given me this opportunity, and will continue to do my best to make you proud.
    [Show full text]
  • Graph Neural Networks for Natural Language Processing: a Survey
    GRAPH NEURAL NETWORKS FOR NLP: A SURVEY Graph Neural Networks for Natural Language Processing: A Survey ∗ Lingfei Wu [email protected] JD.COM Silicon Valley Research Center, USA ∗ Yu Chen [email protected] Rensselaer Polytechnic Institute, USA y Kai Shen [email protected] Zhejiang University, China Xiaojie Guo [email protected] JD.COM Silicon Valley Research Center, USA Hanning Gao [email protected] Central China Normal University, China z Shucheng Li [email protected] Nanjing University, China Jian Pei [email protected] Simon Fraser University, Canada Bo Long [email protected] JD.COM, China Abstract Deep learning has become the dominant approach in coping with various tasks in Natural Language Processing (NLP). Although text inputs are typically represented as a sequence of tokens, there is a rich variety of NLP problems that can be best expressed with a graph structure. As a result, there is a surge of interests in developing new deep learning techniques on graphs for a large number of NLP tasks. In this survey, we present a comprehensive overview on Graph Neural Networks (GNNs) for Natural Language Processing. We propose a new taxonomy of GNNs for NLP, which systematically organizes existing research of GNNs for NLP along three axes: graph construction, graph representation learning, and graph based encoder-decoder models. We further introduce a large number of NLP applications that are exploiting the power of GNNs and summarize the corresponding benchmark datasets, evaluation metrics, and open-source codes. Finally, we discuss various outstanding challenges for making the full use of GNNs for NLP as well as future research arXiv:2106.06090v1 [cs.CL] 10 Jun 2021 directions.
    [Show full text]
  • AI in HR: the New Frontier INTRODUCTION the Robots Have Arrived
    INHR PEOPLE + FOCUSSTRATEGY’S WHITE PAPER SERIES AI in HR: The New Frontier INTRODUCTION The robots have arrived. Only they don’t look like robots. AI- and machine learning-based technology is embedded in the tools you likely interact with every day—navigation apps; predictive text in e-mail and messaging; digital assistants like Alexa and Siri; and recommendation engines on Netflix, YouTube and Spotify. Today, AI-powered technology seamlessly integrates into the things we already do—and it’s poised to do the same for HR. Since AI-based solutions predict, recommend and communicate based on data, AI has the opportunity to transform HR practices in areas where there is adequate data and where that data can be utilized to create greater efficiency, communicate at scale, provide recommendations and predict outcomes. Many companies already have a vast amount of data on candidates and employees that can help them source, assess, hire, train, develop and pay people more efficiently with the application of AI-driven tools. he International Data Corporation smarter hiring choices that ensure quali- T(IDC) estimates that worldwide spend- ty hires while eliminating hidden biases. ing on AI systems will reach $35.8 billion The recruiting process is filled with in 2019 and $79 billion by 2022. And Gart- numerous repeatable tasks such as re- ner estimates that, by 2022, AI-derived sume screening and interview scheduling business value will reach $3.9 trillion. that, when given to AI, free up recruiters’ With its vast amount of underutilized time to focus on what matters: the human data, HR has the potential to lead digital interaction between candidates and com- transformation and drive business value pany personnel.
    [Show full text]
  • Linguistics 1
    Linguistics 1 Purpose of the PhD Program LINGUISTICS The goal of the linguistics doctoral program is to prepare graduates to design and conduct original, empirically based research within a Linguistics is the study of all aspects of human language: how languages theoretical framework. Doctoral students prepare for careers in academic make it possible for us to construe another person's ideas and feelings; research and teaching or applied work in industry or other organizations. how and why languages are similar and different; how we develop We encourage doctoral students to begin engaging in research projects different styles and dialects; what will be required for computers to early, as these will contribute to developing the research program that understand and produce spoken language; and how languages are used will lead to the thesis project. Early projects not only lead to preliminary in everyday communication as well as in formal settings. Linguists exam topics and/or publishable papers, but also enable students to pilot try to figure out what it is that speakers know and do by observing the methods that will be usable for the dissertation project. Early projects structure of languages, the way children learn language, slips of the may be extensions of coursework and may involve a faculty advisor other tongue, conversations, storytelling, the acoustics of sound waves and the than the thesis advisor. way people's brains react when they hear speech or read. Linguists also reconstruct prehistoric languages, and try to deduce the principles behind Doctoral students complete a core of required courses that provide their evolution into the thousands of languages of the world today.
    [Show full text]
  • Corpus-Based Language Technology for Indigenous and Minority Languages
    Corpus-based language technology for indigenous and minority languages Kevin Scannell Saint Louis University 5 October 2015 Background ● I was born in the US, but I speak Irish ● Trained as a mathematician ● Began developing tools for Irish in the 90s ● No corpora for doing statistical NLP ● So I built my own with a crawler ● Crúbadán = crawling thing, from crúb=paw Web as Linguistic Archive ● Billions of unique resources, +1 ● Raw data only, no annotations, -1 ● > 2100 languages represented, +1 ● Very little material in endangered languages, -1 ● Full text searchable, +1 ● Can't search archive by language, -1 ● Snapshots archived periodically, +1 ● Resources removed or changed at random, -1 ● Free to download, +1 ● Usage rights unclear for vast majority of resources, -1 An Crúbadán: History ● First attempt at crawling Irish web, Jan 1999 ● 50M words of Welsh for historical dict., 2004 ● ~150 minority languages, 2004-2007 ● ~450 languages for WAC3, 2007 ● Unfunded through 2011 ● Search for “all” languages, started c. 2011 How many languages? ● Finishing a 3-year NSF project ● Phase one: aggressively seek out new langs ● Phrase two: produce free+usable resources ● Current total: 2140 ● At least 200 more queued for training ● 2500? 3000? Design principles ● Orthographies, not languages ● Labelled by BCP-47 codes ● en, chr, sr-Latn, de-AT, fr-x-nor, el-Latn-x-chat ● Real, running texts (vs. word lists, GILT) ● Get “everything” for small languages ● Large samples for English, French, etc. How we find languages ● Lots of manual web searching!
    [Show full text]
  • Predictive Text System for Bahasa with Frequency, N-Gram, Probability Table and Syntactic Using Grammar
    Predictive Text System for Bahasa with Frequency, n-gram, Probability Table and Syntactic using Grammar Derwin Suhartono, Garry Wong, Polim Kusuma and Silviana Saputra Computer Science Department, Bina Nusantara University, K H. Syahdan 9, Jakarta, Indonesia Keywords: Predictive Text, Word Prediction, n-gram, Prediction, KSPC, Keystrokes Saving. Abstract: Predictive text system is an alternative way to improve human communication, especially in matter of typing. Originally, predictive text system was intended for people who have flaws in verbal and motor. This system is aimed to all people who demands speed and accuracy in typing a document. There were many similar researches which develop this system that had their own strengths and weaknesses. This research attempts to develop the algorithm for predictive text system by combining four methods from previous researches and focus only in Bahasa (Indonesian language). The four methods consist of frequency, n-gram, probability table, and syntactic using grammar. Frequency method is used to rank words based on how many times the words were typed. Probability table is a table designed for storing data such as predefined phrases and trained data. N-gram is used to train data so that it is able to predict the next word based on previous word. And syntactic using grammar will predict the next word based on syntactic relationship between previous word and next word. By using this combination, user can reduce the keystroke up to 59% in which the average keystrokes saving is about 50%. 1 INTRODUCTION verbal and motor (Vitoria and Abascal, 2006). Over time, the usage of this system began to change and Conventional process of typing documents using a now is aimed to all people who demands speed and typewriter has become obsolete due to technological accuracy in typing a document.
    [Show full text]
  • Voice Typing: a New Speech Interaction Model for Dictation on Touchscreen Devices
    Voice Typing: A New Speech Interaction Model for Dictation on Touchscreen Devices †‡ † † Anuj Kumar , Tim Paek , Bongshin Lee † ‡ Microsoft Research Human-Computer Interaction Institute, One Microsoft Way Carnegie Mellon University Redmond, WA 98052, USA Pittsburgh, PA 15213, USA {timpaek, bongshin}@microsoft.com [email protected] [12] and other ergonomic issues such as the “fat finger ABSTRACT problem” [10]. Automatic dictation using speech Dictation using speech recognition could potentially serve recognition could serve as a natural and efficient mode of as an efficient input method for touchscreen devices. input, offering several potential advantages. First, speech However, dictation systems today follow a mentally throughput is reported to be at least three times faster than disruptive speech interaction model: users must first typing on a hardware QWERTY keyboard [3]. Second, formulate utterances and then produce them, as they would compared to other text input methods, such as handwriting with a voice recorder. Because utterances do not get or typing, speech has the greatest flexibility in terms of transcribed until users have finished speaking, the entire screen size. Finally, as touchscreen devices proliferate output appears and users must break their train of thought to throughout the world, speech input is (thus far) widely verify and correct it. In this paper, we introduce Voice considered the only plausible modality for the 800 million Typing, a new speech interaction model where users’ or so non-literate population [26]. utterances are transcribed as they produce them to enable real-time error identification. For fast correction, users However, realizing the potential of dictation critically leverage a marking menu using touch gestures.
    [Show full text]
  • Proceedings of the 55Th Annual Meeting of the Association For
    ComputEL-3 Proceedings of the 3rd Workshop on the Use of Computational Methods in the Study of Endangered Languages Volume 1 (Papers) February 26–27, 2019 Honolulu, Hawai‘i, USA Support: c 2019 The Association for Computational Linguistics Order copies of this and other ACL proceedings from: Association for Computational Linguistics (ACL) 209 N. Eighth Street Stroudsburg, PA 18360 USA Tel: +1-570-476-8006 Fax: +1-570-476-0860 [email protected] ISBN 978-1-950737-18-5 ii Preface These proceedings contain the papers presented at the 3rd Workshop on the Use of Computational Methods in the Study of Endangered languages held in Hawai’i at Manoa,¯ February 26–27, 2019. As the name implies, this is the third workshop held on the topic—the first meeting was co-located with the ACL main conference in Baltimore, Maryland in 2014 and the second one in 2017 was co-located with the 5th International Conference on Language Documentation and Conservation (ICLDC) at the University of Hawai‘i at Manoa.¯ The workshop covers a wide range of topics relevant to the study and documentation of endangered languages, ranging from technical papers on working systems and applications, to reports on community activities with supporting computational components. The purpose of the workshop is to bring together computational researchers, documentary linguists, and people involved with community efforts of language documentation and revitalization to take part in both formal and informal exchanges on how to integrate rapidly evolving language processing methods and tools into efforts of language description, documentation, and revitalization. The organizers are pleased with the range of papers, many of which highlight the importance of interdisciplinary work and interaction between the various communities that the workshop is aimed towards.
    [Show full text]
  • Experts Doubt Ethical AI Design Will Be Broadly Adopted As the Norm Within the Next Decade
    FOR RELEASE June 16, 2021 Experts Doubt Ethical AI Design Will Be Broadly Adopted as the Norm Within the Next Decade A majority worries that the evolution of artificial intelligence by 2030 will continue to be primarily focused on optimizing profits and social control. They also cite the difficulty of achieving consensus about ethics. Many who expect progress say it is not likely within the next decade. Still, a portion celebrate coming AI breakthroughs that will improve life BY Lee Rainie, Janna Anderson and Emily A. Vogels FOR MEDIA OR OTHER INQUIRIES: Lee Rainie, Director, Internet and Technology Research Janna Anderson, Director, Elon University’s Imagining the Internet Center Haley Nolan, Communications Associate 202.419.4372 www.pewresearch.org RECOMMENDED CITATION Pew Research Center, June 16, 2021. “Experts Doubt Ethical AI Design Will Be Broadly Adopted as the Norm in the Next Decade” 1 PEW RESEARCH CENTER About Pew Research Center Pew Research Center is a nonpartisan fact tank that informs the public about the issues, attitudes and trends shaping America and the world. It does not take policy positions. It conducts public opinion polling, demographic research, content analysis and other data-driven social science research. The Center studies U.S. politics and policy; journalism and media; internet, science and technology; religion and public life; Hispanic trends; global attitudes and trends; and U.S. social and demographic trends. All of the center’s reports are available at www.pewresearch.org. Pew Research Center is a subsidiary of The Pew Charitable Trusts, its primary funder. For this project, Pew Research Center worked with Elon University’s Imagining the Internet Center, which helped conceive the research and collect and analyze the data.
    [Show full text]